Matched time-alignment profiles for an unresolved stretch

The saved fits for stretch 0e73879f98c3b701fd7c42df disagree on the time alignment between IMU increments and receiver positions. The zero-field globe offsets differ by about 17.3 milliseconds without upstream field removal and 70.0 milliseconds when removal is fitted. Timing changes can be partly absorbed by initial-position, acceleration and other permitted calibration adjustments.

This diagnostic holds each of those four exact saved offsets fixed and fits all other active parameters twice: from the saved primary solution and from its saved repeat. Both processing cases are included. Two fresh smooth controls run first; failed controls prevent target starts.

Implementation and checks

analysis/tests/ilvis0_remaining_time_profile_worker.py prepares, freezes and runs the new study. It privately adapts the completed fixed-time executor; that executor and its studies remain unchanged.
analysis/tests/test_ilvis0_remaining_time_profile_worker.py checks both processing cases, both starts, the already-passing source repeat, exact offsets, parameter order, invalid populations and template isolation.
The existing lll.ilvis0_fixed_time_profile.FixedTimeProblem and bounded trust solver retain their physical bounds and stopping tolerances. Existing adapter controls include three compiled-motion known-answer cases.
The fresh full-Jacobian diagnostic uses the receipt ledger's supported tangent call kind. The earlier study's accounting failure remains recorded.
Final artifacts retain full parameters, residuals, Jacobians, timing derivative, observation epochs and receiver geometry. Raw source hashes, extraction provenance, source snapshots and environment are checked.

Finite scope

Item	Limit
Fresh smooth controls	2 identities × 3 native calls
Conditional nuisance fits	8 identities × 80 fitting calls
Final full-Jacobian diagnostics	1 per nuisance fit
Total charged identities	10
Total native calls	654 maximum
Saved within-anchor comparisons	4
Numerical threads	4 total, one worker
Retries	0

At the recent rate of approximately 22 seconds per native call, exhausting the allowance would take about four hours. A conditional fit can stop much earlier. No completed allowance is reused or extended. Interrupted identities remain charged; incomplete receipt transactions block resumption.

Required interpretation

Report every offset, both starts and every failure. At each common fixed offset, compare conditional completion, relative objective difference and reconstructed track agreement under the existing 10⁻⁶ cost / 0.01 m axis-RMS limits.

Fixed-time completion does not demonstrate free-time convergence. These results cannot change the public shape tally. Any later full comparison must pass its original primary, repeat and companion requirements under a separate finite freeze. Measurements, selected interval, candidate models, receiver weights, nominal integration clock and physical bounds remain unchanged.

The other two unresolved stretches retain their separate timing/calibration diagnoses. This study neither expands to them nor starts an Earth-model campaign.